CpG methylation signature predicts prognosis in breast cancer.

Abstract

DNA methylation can be used as prognostic biomarkers in various types of cancers. We aimed to identify a CpG methylation pattern for breast cancer. In this study, using the microarray data from the cancer genome atlas (TCGA) and gene expression omnibus (GEO), we profiled DNA methylation between 97 healthy control samples and 786 breast cancer samples in a training cohort (from TCGA, n = 883) to build a gene classifier using a penalized regression model. We validated the prognostic accuracy of this gene classifier in an internal validation cohort (from GEO, n = 72). A total of 1777 differentially methylated CpGs corresponding to 1777 different methylated genes (DMGs) between breast cancer and control were chosen for this study. Subsequently, 16 CpGs were generated to classify patients into high-risk and low-risk groups in the training cohort. Patients with high-risk scores in the training cohort had shorter overall survival (hazard ratio [HR], 4.674; 95% CI 2.918 to 7.487; P = 1.678e-12) than patients with low-risk scores. The prognostic accuracy was also validated in the validation cohorts. Furthermore, among patients with low-risk scores in the combined training and validation cohorts, the patients with the age > 60years compared with the patients with the age < 60years were associated with improved overall survival (HR 2.088, 95% CI 1.348 to 3.235; p = 7.575e-04) in patients with a high-risk score but not in patients with low-risk score (HR 1.246, 95% CI 0.515 to 3.011; p = 0.625). The patients treated with radiotherapy compared with the patients without radiotherapy were associated with improved overall survival (HR 0.418, 95% CI 0.249 to 0.703; p = 6.991e-04) in patients with a high-risk score but not in patients with low-risk score (HR 2.092, 95% CI 0.574 to 7.629; p = 0.253). For the patients with recurrence and the patients without recurrence both groups were all associated with improved overall survival (HR 7.475, 95% CI 4.333 to 12.901; p = 6.991e-04) in patients with a high-risk score and in patients with low-risk score (HR 14.33, 95% CI 4.265 to 48.17; p = 4.883e-13). The 16 CpG-based signature is useful as a biomarker in predicting prognosis for patients with breast cancer.